AI coding tools are changing how programmers work, but no study we read shows them replacing the profession. As of 24 October 2026, the US Bureau of Labor Statistics projects software developer jobs to grow 10 percent from 2025 to 2035 and computer programmer jobs to fall 7 percent. Controlled studies of speed disagree, from large gains to a slowdown, so the answer depends on the task, the tool and the developer.
01What the US job numbers say for coders
The US Bureau of Labor Statistics (BLS) counted about 1.9 million jobs in the combined group of software developers, quality assurance analysts and testers in 2025 and projects growth of 10 percent to 2035, much faster than average. It ties the demand partly to continued expansion of software built for AI. For computer programmers it projects a 7 percent fall from 110,800 jobs, saying programming work continues to be automated and some higher-skilled tasks may move to developers.
Both occupations are rated Very high for AI exposure in the BLS exposure table. In 2025 case studies BLS reasoned that AI could raise developer productivity and lower prices, which might raise demand, and it saw no clear evidence for a decline. That reasoning used information from June 2024, and BLS says its methods assume technology moves at past speeds and are not built to capture extremely rapid change.
02Four studies of speed and what each can show
Studies measure different things, and none covers all coding work. Read the Copilot and company results as evidence of gains on defined tasks, and the METR trial as evidence about experienced people working on code they know well.
- GitHub Copilot experiment, 2023. Recruited developers were asked to write a web server in JavaScript. The group with the assistant finished 55.8 percent faster. Limit: one self-contained task.
- Three company field experiments. At Microsoft, Accenture and a Fortune 100 company, 4,867 developers were randomly given an AI assistant or not. Pooled, the AI group completed about 26 percent more tasks, with a wide margin of error. Less experienced developers adopted the tool more and gained more.
- METR trial, early 2025. Sixteen experienced open-source developers worked on 246 real tasks in their own large projects, with AI allowed or barred at random. With AI, tasks took 19 percent longer. The developers had expected to be 24 percent faster, and afterwards still believed they had been about 20 percent faster.
- METR update, February 2026. A later experiment pointed to time savings of about 18 percent for returning developers and 4 percent for new ones, but both ranges included zero. METR called the signal unreliable because many developers refused to work without AI.
03Why the results differ
The studies used different tasks, from a single built-from-scratch exercise to real fixes in familiar codebases. They used different developers, from recruits to veteran maintainers, and different tools, from early code-completion assistants to agentic tools. They also measured different outcomes, such as time to finish or tasks completed. METR's own limits are plain: its early-2025 result covers one setting and a small group, and it says it does not show that AI slows most developers. After its 2026 update METR said developers are probably more sped up now than in early 2025, but called its data very weak evidence for the size of that change.
Faster individual coding does not settle how many coders are hired. BLS reasoned that cheaper software could raise demand, and nobody has yet measured whether that outweighs the time saved on each task.
04What it means at the start of a career
The signals about beginners point two ways. In the company field experiments, less experienced developers gained more from the assistant. Yet a Stanford payroll study, revised on 12 August 2026, finds employment of workers aged 22 to 25 in AI-exposed occupations is 19 percent below where it would be had they tracked less-exposed jobs, mainly through less hiring. It is not a study of coders alone, since the gap also holds when computer occupations are excluded, and the authors call it descriptive, not causal.
This page is general information, not career advice. For decisions about training or work, ask a qualified careers adviser.
?What is not known
- How much faster AI makes working developers today is not settled. The best randomised evidence is small, and METR says its latest data is unreliable.
- Whether weaker hiring of beginners in exposed jobs is caused by AI, or by the wider economy, is disputed.
- Newer agentic tools arrived after most of these studies were run, so their effect on jobs is largely unmeasured.
QQuestions people ask
Is a software developer the same as a programmer?
BLS counts them as separate occupations with different outlooks: developers are projected to grow 10 percent and programmers to fall 7 percent. Its programmer page notes that some higher-skilled programming tasks may shift to developers.
Why did developers in the METR trial misjudge their own speed?
The study does not explain it. It only reports that developers expected a speedup before and still believed they had one afterwards, which is why METR treats self-reports as unreliable.
Last checked October 24, 2026. Rules and settings change, so open the source to see the current version.
- Software Developers, Quality Assurance Analysts, and Testers, Occupational Outlook Handbook US Bureau of Labor Statistics Used for 2025 employment, projected growth to 2035 and the stated link to AI software demand.
- Computer Programmers, Occupational Outlook Handbook US Bureau of Labor Statistics Used for the projected decline and the handbook's text on automating programming.
- AI exposure categories and 2025-35 projections (table) US Bureau of Labor Statistics Occupation-level table used for the Very high exposure rating of developers and programmers.
- Incorporating AI impacts in BLS employment projections: occupational case studies US Bureau of Labor Statistics (Monthly Labor Review) February 2025 article used for BLS's reasoning on software developers and its June 2024 information date.
- The Impact of AI on Developer Productivity: Evidence from GitHub Copilot arXiv 2023 controlled experiment used for the 55.8 percent faster result on one task.
- The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers Microsoft Research Research page used for the pooled field experiment result and the finding on less experienced developers.
- Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity METR Randomised trial write-up used for the 19 percent slowdown, the developers' expectations and METR's stated limits.
- We are Changing our Developer Productivity Experiment Design METR February 2026 update used for the later estimates, their ranges and METR's reasons for doubting them.
- Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI Stanford Digital Economy Lab Payroll-data working paper used for the finding on workers aged 22 to 25 and its caveats.
General information, not legal, medical, financial or investment advice. We are not affiliated with any company, agency or standards body named here. See our editorial policy and corrections, or tell us about a mistake at [email protected].